Which salary data should you trust?
Surveys, live data and job ads answer different questions. Pick the source for each job before you see its number.
Short answer: Trust each source for the question it answers. Salary surveys are the most reliable base for a pay structure because their job matches are controlled. Live salary data is a useful check on roles whose market moves fast. Job ads show what candidates will compare an offer with, not what the market pays. Decide which source leads for each job family in advance.
Three numbers for one job
Northwind Analytics has 160 people and is building its first proper salary bands. The head of people starts with one role that keeps coming up in exits: Senior Data Engineer. She looks at three sources and gets three answers.
The salary survey the company joined last spring puts the median base pay for the matched job at $162,000. A live compensation platform the finance team trialed shows $178,000 for "Senior Data Engineer, L4". Twelve job ads for similar roles, collected by a recruiter, have a median midpoint of $185,000.
The CFO asks the obvious question: Which one is the market? The recruiter says the job ads, because "that's what candidates see". The finance lead says the live platform, because "it's current". The head of people trusts the survey, because she knows how its matches were made.
All three are partly right. The numbers differ because each source measures something different. The survey shows what a defined group of employers paid people in a matched job on a fixed date. The live platform shows what that platform's customers pay now, matched by those customers. The job ads show what employers say they will pay a new hire. None of them is wrong. They answer different questions.
Northwind Analytics is a fictional company and the figures are illustrative.
Why choosing salary data is harder than it looks
Each kind of market data trades one virtue for another. Knowing what each one gives up is most of the skill.
Salary surveys trade speed for control. A traditional survey asks participating employers to match their jobs to written descriptions and submit pay for the people in them. Results are cleaned, aggregated and published for one effective date. That process is slow by design. Even the US government's own occupational wage survey combines data collected over three years and adjusts wages from earlier panels forward to a single May reference date [1]. Commercial surveys are faster than that, but the data in them is usually months old on the day it arrives.
Live data trades control for speed. Platforms that read pay straight from their customers' HR and payroll systems can update continuously. The trade-off is who is inside the sample and how jobs are matched. The sample is whoever buys the platform. Jobs are usually mapped by those customers to the platform's own levels, with less review than a survey team applies.
"Live" also has limits that are easy to miss. Pay for most people changes only twice: When they are hired and at the annual review. So even a perfectly current feed shows a market that mostly moved months ago, plus the offers made since. What it adds is those recent offers. That matters a great deal for a skill in short supply and very little for a role whose pay moves with the yearly budget.
Job ads trade accuracy for reach. Posted pay is now visible for about half of US job ads. A job site's public tracker shows 51.8% of US postings on its site with explicit pay information in August 2026, up from about 14% in January 2020 [2]. Postings are free and current. But they show what employers advertise, not what they pay. Research from the Minneapolis Fed finds that average posted wages run about 40% above the wages of workers in government data, with the widest gap at the bottom of the pay scale [3].
Underneath is the tension every compensation team lives with. The first goal is credibility: A number you can defend to a CFO, an employee or a regulator, because you know where it came from. The second is currency: A number that reflects the market a candidate is facing this month. The sources that score best on one tend to score worst on the other.
Five ways teams misread their market data
Each of these looks sensible at a growing company. Each produces bands that are hard to explain later.
"The freshest number is the truest"
What it looks like: The team treats the most recent number as the most accurate one and lets it override the survey wherever they differ.
Why smart teams fall into it: Old data feels wrong in a moving market. Live dashboards look precise and refresh constantly. A number from last spring looks stale next to one from last week.
What it costs: Fresh is not the same as representative. A live sample reflects who uses the platform. If its customers are mostly venture-backed software companies, its numbers describe that market well and others poorly. A 160-person analytics firm in Denver may or may not be in that peer group. For a support manager at a logistics company, the live sample may hold only a handful of matches. The number still looks precise. The precision just isn't real.
There is a second, quieter problem: Sample drift. As customers join and leave a platform, the group behind each number changes from month to month. A rising line can mean the market moved. It can also mean that higher-paying companies signed up. A survey has the same risk from one year to the next, but it is published with the participant list or at least its profile, so the change can be seen. A live line rarely says who left the sample last quarter.
"Job ads are free market data now"
What it looks like: With so many ranges now public, a recruiter collects a dozen ads and uses their midpoints as the benchmark.
Why smart teams fall into it: It costs nothing. It is what candidates actually see. And pay transparency laws were meant to make this kind of comparison possible.
What it costs: Posted ranges are offers in the making, not pay. They are often wide: The Minneapolis Fed study finds that ranges in posts span about 28% of their midpoint on average [3]. Which firms post is not random either. In the same study, high-wage firms and occupations were less likely to post pay and posted wider ranges when they did [3]. Coverage also varies by job family. In August 2026, the same tracker showed pay on 49.1% of US software development postings, against 82.5% for childcare [2]. In Europe the gaps are larger. Germany sat at 14.7% and the UK at 59.5% in the same month [2]. A dozen ads tell you what a dozen employers chose to advertise. They do not tell you where the market pays.
Posted data has a real use. It shows what a candidate will compare your offer to. That makes it a check on the offer, not a basis for the band.
"Blending the sources averages out the error"
What it looks like: Faced with three numbers, the team averages them. $162,000, $178,000 and $185,000 become a "blended market rate" of about $175,000.
Why smart teams fall into it: Blending is standard practice in compensation work and it feels balanced. Nobody has to choose a winner.
What it costs: A blend is only sound when its parts measure the same thing. These three do not. They come from different dates, different populations and different definitions of the job. One of them is not even pay. Averaging them produces a number that belongs to no market and that no one can explain. Blending works when sources share an effective date, a pay definition and a match you trust. It fails when it is used to avoid deciding which source fits.
Dates are the most common hidden difference. A survey with an effective date of last April and a live number from this month are a year apart before any other difference is counted. Bringing both to one date is routine work. Deciding how fast to move the older number forward is a judgment about how the market has behaved since. In a quiet year the choice barely matters. For a hot skill it can be worth more than the gap between the sources.
"One source for every role"
What it looks like: The company picks one data provider and uses it for every job, from accounts payable clerk to machine learning engineer.
Why smart teams fall into it: One source is cheaper, simpler and consistent. Mixing sources feels like an invitation to cherry-pick.
What it costs: Sources are strong in different places. A broad survey with deep coverage is reliable for finance, HR, operations and customer support, where jobs are well defined and pay moves slowly. It can lag badly for a skill whose market changed in a quarter. A live source can track that fast-moving role, but may have thin data for a warehouse lead or a payroll specialist. Using one source everywhere means accepting its weakest areas as part of the structure. The consistency is real. So are the blind spots.
"We'll pick the source once we see the numbers"
What it looks like: For each role, the team looks at all its sources and uses whichever looks right. The survey for one job, the live data for the next, the job ads for a third.
Why smart teams fall into it: Every individual choice can be justified. This survey match is weak, that live sample is small. Experienced people are trusted to make the call.
What it costs: Over a whole structure, the pattern shows. Roles a manager is fighting for get the highest source. Roles nobody is watching get the lowest. Pay for similar work drifts apart and nobody can say why, except that different sources were used. That is hard to explain to an employee who asks how their pay was set. It is harder still in the EU, where workers can ask for the criteria behind pay and pay progression [6]. A source rule set in advance looks dull. A source picked per role looks like a decision made after the answer was known.
What a sound market data policy looks like
There is no best source. There is a sound way to use several. In our view it has these qualities.
- Each job family has a primary source, chosen in advance. The choice follows coverage and the pace of the market for that family, not the number it produces. It is written down and reviewed once a year.
- Every number has a known date. Data from different sources is brought to one effective date before it is compared. A survey number is never set beside a live number without saying how old each is.
- The sample is known for every cut. For each benchmark the team knows roughly how many companies and people sit behind it and what kind of companies they are. A cut with too few data points is flagged rather than used.
- Live data is a check, not an override. When a live source and the primary source disagree by more than a stated margin, someone looks again at the match and the sample. They do not simply take the higher number.
- Fast-moving roles have a named review date. The few roles where the market moves inside a year are listed, with a date when their numbers are looked at again. Everyone else waits for the annual cycle.
- Job ads inform offers, not bands. Posted ranges show what candidates will compare an offer with. They feed the offer conversation and the wording of your own ads. They do not set the structure.
- The match matters more than the source. A strong source with a weak match is worse than a modest source with a strong one. Our guide on whether market data can set your salary bands looks at matching and peer groups in depth.
- The policy can be explained in a paragraph. If a manager or employee asks where a number came from, the answer is the same every time.
None of this is a method to copy. Each quality is a test a reviewer can put to a company's own benchmarks.
The calls only you can make
Choosing data is partly technical. Several of the choices behind it are not.
Which market you compete in. The companies you lose people to and the companies you look like are often different. A data source answers only for the market you point it at. Pointing it is your call.
How much to pay for speed. Faster data costs more, in money and in the effort to check it. For a company where most roles move slowly, that may not be worth it. For one built around a skill in short supply, it may be essential.
Whether a spike is a premium or the new rate. When live data jumps for one skill, the company has to decide whether to pay it as a time-limited premium or move the band. That is a judgment about the market's future, not a reading of its present.
Whether to take part in surveys. Most surveys give their full results only to employers that submit data. Taking part costs real effort each year: Matching jobs, cleaning data and answering queries. It also gives the company a seat in the market it measures. Staying out saves the effort but leaves the company buying whatever is sold to non-participants, often less detail at a higher price.
How open to be about sources. Some companies tell staff which data they use. Others say only that bands are market-based. With more pay laws asking for criteria and good-faith ranges, the more open answer is getting easier to defend and the vague one harder.
The data and the numbers
A few dated facts frame the choice.
- Posted pay is spreading, but slowly now. The tracker puts US postings with pay at 51.8% in August 2026 [2]. Its European data shows wide differences by country: In March 2026, 48% of Dutch postings and 12% of German ones showed pay [4]. Ranges also differ in width: The median hourly range in the UK was about 11% of its lower bound, against about 50% in Italy [4].
- Different measures give different answers. The Minneapolis Fed study finds that only 14% of the online posts it studied contain any pay information [3]. Its posts and the tracker's come from different sources and periods. That is the point: Even the share of ads with pay depends on whose data you read.
- Posting laws have moved posted pay. A 2025 study of state posting laws found wages rose 1.3% to 3.6% after the laws took effect, with no effect on pay dispersion, employment or the number of postings [5]. What candidates see now shapes what employers pay.
- The EU will add more posted data, not less. The EU directive gives applicants the right to pay information before the interview, though not necessarily in the ad [6] [4]. As it reaches national law, more European pay data will become visible to candidates.
- Government surveys run on long cycles. US occupational wage estimates combine three years of survey panels, aged to one reference date [1]. They are reliable for broad occupations and too slow and coarse for setting one company's bands.
Our view
We think the question "survey or live data?" is the wrong one. The right one is "which source answers which question, for which jobs?"
For most companies of 50 to 500 people, a broad survey with strong job matching should carry the structure. Live data earns its place as a check on the few roles whose market is moving fast. Job ads belong in the offer conversation, where they describe what a candidate sees. They belong nowhere near the bands.
We would argue hardest against two habits: Averaging sources that measure different things and choosing a source after seeing what it says. Both produce bands that look market-based and cannot be explained. A source rule written down before the numbers arrive is the simplest protection a pay structure has.
Where we help
Our market pricing work sets a source rule for each job family with you, checks the matches behind every benchmark and brings your data to one date, so each number in your structure can be explained. The starting point is a market pricing project, quoted within one business day. If a conversation would help first, Book a free call.
*We don't provide market data. We use the data you have or buy. No data yet? We help you choose a survey or tool that fits your size and budget.
Sources
- US Bureau of Labor Statistics, Occupational Employment and Wage Statistics: Survey methods (page modified 15 May 2026), checked 2026-10-25
- Indeed Hiring Lab, Pay transparency tracker data (country and sector series, through August 2026), checked 2026-10-25
- Batra, Michaud and Mongey, Online job posts contain very little wage information, Federal Reserve Bank of Minneapolis Staff Report 690 (28 September 2026), checked 2026-10-25
- Indeed Hiring Lab, Full salary transparency in Europe is still a distant prospect (7 May 2026), checked 2026-10-25
- Arnold, Quach and Taska, The impact of pay transparency in job postings on the labor market, NBER Working Paper 34480 (November 2025), checked 2026-10-25
- Directive (EU) 2023/970 on pay transparency, Articles 5 and 6 (EUR-Lex), checked 2026-10-25